Facial prediction model construction method, facial prediction method and related device

Abstract

The application relates to the technical field of face prediction, and is a face prediction model construction method, a face prediction method and related devices.The method comprises the following steps: acquiring a training set, a test set and a verification set; each sample in the training set, the test set and the verification set comprises genomics features and real facial morphology features; a training number is set; a preset artificial neural network model is trained multiple times according to the training number and the training set; a plurality of face prediction standby models corresponding to the training are obtained; the verification set is predicted by using the obtained plurality of face prediction standby models; the optimal parameters are selected; and a face prediction model is output. Compared with the prior art, the application is cleaner in sample extraction, the sample quantity is improved, the model prediction obtained through training is more accurate, and furthermore, the application adopts neural network modeling, and the test set and the verification set are used for evaluating and analyzing the face prediction model, so that the prediction accuracy and precision of the model are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of face prediction, and is a face prediction model construction method, a face prediction method and related devices. BACKGROUND

[0002] Facial features are the identity recognition cards of human beings from birth, and have a very high individual recognition degree. Compared with eye color, hair features, height and body shape, and other physical features, facial features are not easy to modify and change, and are less affected by environmental factors, so they have higher application value in identity recognition. At the same time, research shows that facial features have a high degree of genetic conservation. This means that the differences in individual facial features are mainly determined by genetic factors, and when we analyze enough genetic codes of facial features, we can make very accurate predictions about faces. However, there are few studies on predicting three-dimensional facial features based on genomic information.

[0003] Existing research mainly constructs algorithm models based on SNP site data associated with appearance features. Due to the high complexity of facial features, sample size and other factors, not many genetic sites have been found, and a large number of sites have not been discovered, let alone used for face model establishment. SUMMARY

[0004] The present application provides a face prediction model construction method, a face prediction method and related devices for predicting faces based on genetic data.

[0005] Therefore, the first aspect of the present application provides a face prediction model construction method, comprising:

[0006] Obtaining a training set, a test set and a validation set, wherein each sample in the training set, the test set and the validation set includes genomic features and real facial morphological features;

[0007] Setting a training number, and training a pre-set artificial neural network model multiple times using the training set according to the training number to obtain a plurality of face prediction standby models;

[0008] Using the obtained plurality of face prediction standby models to predict the validation set to obtain corresponding loss function values, adjusting the model parameters according to the loss function values in the validation set, selecting the optimal parameters, and outputting the optimal face prediction standby model as a face prediction model.

[0009] The above obtaining a training set, a test set and a validation set comprises:

[0010] Determining a plurality of samples, and dividing all samples into a training set, a test set and a validation set, wherein each sample includes corresponding genetic SNP site data and real facial morphological features;

[0011] Perform principal component analysis on all genetic SNPs site data in the training set to obtain a new feature space, select the 97% principal components in the new feature space of each sample as the genomic feature of the sample, and obtain the real facial morphological feature of each sample in the same way;

[0012] Map all genetic SNPs site data in the test set and the validation set to the new feature space of the training set, select the 97% principal components in each sample as the genomic feature of the sample, and obtain the real facial morphological feature of each sample in the same way.

[0013] The acquisition step of the SNPs site data of each sample includes:

[0014] Extract the venous blood of the sample, and obtain the genetic SNPs site data by sequencing typing;

[0015] Pretreat the genetic SNPs site data, and output the genetic SNPs site data of each sample, wherein the pretreatment includes cleaning and quality control.

[0016] The acquisition step of the real facial morphological feature of each sample includes:

[0017] Perform face scanning on the sample to obtain a three-dimensional face image;

[0018] Obtain the original face point set data corresponding to the three-dimensional face image;

[0019] Automatically perform non-rigid registration on the original face point set data to obtain the corresponding real facial morphological feature.

[0020] The evaluation of the facial prediction model by using the test set and the validation set includes:

[0021] Input the genomic feature of each sample in the test set into the facial prediction model to output the predicted facial morphological feature;

[0022] Determine the angle value between the predicted facial morphological feature and the real facial morphological feature;

[0023] The validation set performs the same process as the test set;

[0024] Determine the mean value of all angle values corresponding to the test set and the validation set;

[0025] Evaluate the facial prediction model by using the mean value.

[0026] Therefore, the second aspect of the present application provides a facial prediction method, which includes:

[0027] Collect the genomic feature of the user to be tested;

[0028] inputting the genomics features of the to-be-tested user into the face prediction model to obtain predicted face morphological features, and obtaining a face prediction result; wherein the face prediction model is a face prediction model constructed by using a face prediction model construction method.

[0029] Therefore, the third aspect of the present application provides a face prediction model construction device, which comprises:

[0030] a preprocessing unit configured to obtain a training set, a test set and a validation set, wherein the training set, the test set and the validation set each comprise a plurality of genomics features and a plurality of real face morphological features, and the plurality of genomics features and the plurality of real face morphological features correspond to each other in a one-to-one manner;

[0031] a modeling unit configured to set a training number, and train a preset artificial neural network model using the training set for a plurality of times according to the training number to obtain a plurality of face prediction standby models corresponding to the training number;

[0032] a first evaluation unit configured to use the plurality of obtained face prediction standby models to predict the validation set to obtain a corresponding loss function value, adjust model parameters according to the loss function value in the validation set, select optimal parameters, and output an optimal face prediction standby model as a face prediction model;

[0033] a second evaluation unit configured to evaluate the face prediction model using the test set and the validation set.

[0034] Therefore, the fourth aspect of the present application provides a face prediction device, which comprises:

[0035] a collection and processing unit configured to collect genomics features of a to-be-tested user;

[0036] a communication unit configured to input the target face data into the face prediction model;

[0037] a prediction processing module configured to obtain a face prediction result using the predicted face morphological features; wherein the face prediction model is a face prediction model constructed by the face prediction model construction device.

[0038] Therefore, the fifth aspect of the present application provides a face prediction device, which comprises a processor, a memory, and one or more programs stored in the memory and configured to be executed by the processor, the programs comprising instructions for performing steps in a face prediction method.

[0039] Beneficial effects: the present application uses genomics features for model learning, and compared with the existing algorithm model technology mainly based on SNP site data associated with appearance features, the present application extracts a cleaner sample, and improves the sample size, so that the model obtained by training is more accurate in prediction; further, the present application adopts neural network modeling, and uses the test set and the validation set to evaluate and analyze the face prediction model, so that the prediction accuracy and precision of the model are improved. BRIEF DESCRIPTION OF DRAWINGS

[0040] BRIEF DESCRIPTION OF DRAWINGS Figure 1 A model construction flowchart for an embodiment of the present application.

[0041] BRIEF DESCRIPTION OF DRAWINGS Figure 2 A predicted face comparison graph for a validation set in an embodiment of the present application.

[0042] BRIEF DESCRIPTION OF DRAWINGS Figure 3 A predicted face comparison graph for a test set in an embodiment of the present application.

[0043] BRIEF DESCRIPTION OF DRAWINGS Figure 4 An efficacy evaluation graph for an embodiment of the present application.

[0044] BRIEF DESCRIPTION OF DRAWINGS Figure 5 A prediction flowchart for an embodiment of the present application.

[0045] BRIEF DESCRIPTION OF DRAWINGS Figure 6 A face prediction model construction device structure schematic diagram for an embodiment of the present application.

[0046] BRIEF DESCRIPTION OF DRAWINGS Figure 7 A face prediction device structure schematic diagram for an embodiment of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0048] The present application provides a face prediction model construction method, a face prediction method and related devices thereof, to solve the technical problem that the existing face prediction model is mainly based on SNP site data associated with facial appearance features, but the amount of SNP site data associated with facial appearance features is limited, and the complexity is very complex, and the model accuracy is not high.

[0049] In the following, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0050] The embodiments of the present application will be described in detail in the following.

[0051] Embodiment 1: As shown in the accompanying drawings, the present embodiment discloses a face prediction model construction method, comprising: Figure 1

[0052] Step S101, obtaining a training set, a test set and a validation set, wherein each sample in the training set, the test set and the validation set comprises genomic features and real facial morphological features.

[0053] In the embodiments of the present application, the training set, the test set and the validation set are obtained, comprising:

[0054] 1. Determine a plurality of samples, and divide all the samples into a training set, a test set and a validation set, wherein each sample comprises corresponding genetic SNPs site data and real facial morphological features;

[0055] Herein, the obtaining step of the genetic SNPs site data of each sample comprises:

[0056] A. Extracting venous blood of the sample, and obtaining genetic SNPs site data through sequencing typing; here, the genetic SNPs site data can be obtained through GSA chip sequencing typing;

[0057] B. Preprocessing the genetic SNPs site data, and outputting the genetic SNPs site data of each sample, wherein the preprocessing comprises cleaning and quality control; the preprocessing process can be performed through PLINK v1.07 biological information software.

[0058] Herein, the obtaining step of the real facial morphological features of each sample comprises:

[0059] A. Performing face scanning on the sample to obtain a three-dimensional face image; here, a color handheld three-dimensional scanner (Artec Spider, Germany) is used to obtain three-dimensional face data, and the scanning sequence is from one side of the ear of the collected person, moving up and down S-shaped through the nose and face to the other side of the ear, then collecting the image of the lower jaw occluded part downward, and finally returning to the starting position to complete the scanning (single frame curved surface range 90mm×70mm to 180mm×140mm).

[0060] ​B. Obtain the original facial point set data corresponding to the 3D face image; here, the 3D face image can be processed using the Artec Studio 11 Professional software that comes with the 3D scanner to obtain the original facial point set data.

[0061] C. Perform automatic non-rigid registration on the original facial point set data to obtain the corresponding real facial morphological features. Here, the FARS software and 3dMD patient software are used to complete the fully automatic non-rigid registration process. If the quality of the three-dimensional facial image of the sample is poor, the sample will be removed.

[0062] 2. Principal component analysis was performed on all genetic SNP loci data in the training set to obtain a new feature space. 97% of the principal components in the new feature space of each sample were selected as the genomic features of that sample, and the true facial morphological features of each sample were obtained in the same way.

[0063] Principal Component Analysis (PCA) is a statistical method for dimensionality reduction. It uses an orthogonal transformation to convert the original random vector whose components are correlated into a new random vector whose components are uncorrelated, that is, to map n-dimensional features onto k dimensions (k < n).

[0064] 3. Map all genetic SNP locus data from the test and validation sets to the new feature space of the training set. Select 97% of the principal components in each sample as the genomic features of that sample, and obtain the true facial morphological features of each sample in the same way. Here, mapping all genetic SNP locus data from the test and validation sets to the new feature space of the training set directly reduces the dimensionality of the samples in the test and validation sets.

[0065] Step S102: Set the number of training iterations, and train the preset artificial neural network model multiple times using the training set according to the number of training iterations to obtain multiple corresponding face prediction backup models.

[0066] Here, the pre-defined artificial neural network model is trained using a training set based on the number of training iterations. Specifically, in this paper, the genomic features of each sample are used as the input layer, and its corresponding facial morphology is used as the output layer. There are eight hidden layers between the input and output layers. The structure of the hidden layers is shown in Table 1. The input layer (genomic features of a sample) is denoted as X, where X = (x1,...,xd). The k-th hidden layer is denoted as Hk, where Hk = (hk1,...,hkp). q represents the number of neurons in the hidden layer, and the q value can be different for each hidden layer. The output layer (facial morphology features of a sample) is denoted as Y, where Y = (y1,...,ym). Furthermore, the activation function in this network structure is the Tanh function as shown below.

[0067] .

[0068] The value of the jth neuron of the first layer of the hidden layer is: .

[0069] The value of the jth neuron of the k+1th layer of the hidden layer is: .

[0070] The value of the jth neuron of the output layer is: .

[0071] Here, the initial parameter values of the model can be as shown in Table 2.

[0072] In step S103, the obtained plurality of face prediction backup models are used to predict the verification set, to obtain corresponding loss function values. According to the loss function values in the verification set, the model parameters are adjusted, the optimal parameters are selected, and the optimal face prediction backup model is output as the face prediction model.

[0073] Here, the loss function is the sum of the weighted loss functions of the three kinds of data, and the calculation formula is as follows: .

[0074] The three kinds of data are three representations of the predicted facial morphological features output by the model, i.e., the original high-dimensional vector, the principal component vector after principal component analysis, and the principal component vector after standardization.

[0075] In step S104, the face prediction model is evaluated using the test set and the verification set.

[0076] The embodiments of the present application evaluate the face prediction model using the test set and the verification set, including:

[0077] 1. Input the genomic features of each sample in the test set into the face prediction model, and output the predicted facial morphological features;

[0078] 2. Determine the angle value between the predicted facial morphological features and the real facial morphological features;

[0079] Here, the angle value between the predicted facial morphological features and the real facial morphological features is determined, i.e., the SSA angle is calculated, and the calculation formula is as follows: ;

[0080] 3. The verification set performs the same process as the test set;

[0081] 4. Determine the mean value of all angle values corresponding to the test set and the verification set;

[0082] 5. The mean value is used to evaluate the face prediction model. The evaluation can be a set of qualified conditions, the mean value is compared with the qualified conditions, it is judged whether the qualified conditions are met, and in response to not meeting, the model parameters are re-adjusted and trained. In the present application, the verification set is used for model selection and parameter adjustment, the test set is used to evaluate the generalization ability of the model, and the optimal face prediction model is evaluated.

[0083] In embodiment 2, male samples are selected, after data cleaning and quality control, a total of 597 samples are used for subsequent analysis, the age changes from 16 to 59 years old, and the samples are randomly divided into three parts: 420 for training set, 100 for verification set and 77 for test set.

[0084] The training times are set to 1600, the face prediction model is constructed according to the steps of embodiment 1, and is trained for 1600 rounds. After training, the model with the minimum loss function in the verification set is selected and saved as the final face prediction model. The verification set and the test set are respectively brought into the face prediction model to obtain the predicted face morphology characteristics, and the comparison between the predicted face morphology characteristics and the real face morphology characteristics is shown in the comparison chart of the predicted face morphology characteristics and the real face morphology characteristics. Figure 2 、 3 The test set and the verification set are used to evaluate the face prediction model, the similarity curve between the real face morphology characteristics and the predicted face morphology characteristics in the test set and the verification set is obtained, and the random predicted face with the same sample size is randomly extracted. After calculating the angle value between the random predicted face and the real face, the mean value is obtained. After repeating the random sampling and calculation process 1000 times, the 1000 random similarity statistical quantities are formed. The similarity curve between the real face morphology characteristics and the random predicted output face morphology characteristics is shown in the similarity curve between the real face morphology characteristics and the random predicted output face morphology characteristics. Figure 4 The P value of the one-sided detection of the normal distribution is used to evaluate the prediction effect of the present application. It can be analyzed that in the verification set, the similarity between the real face and the predicted face of the present application is significantly higher than that of the random predicted face (the value is 2.52x10-16, less than 0.05); in the test set, the similarity between the real face and the corresponding predicted face is also significantly higher than that of the random face (the value is 9.99x10-5, less than 0.05).

[0085] In summary, it can be seen that the face prediction model constructed by the face prediction model construction method disclosed in the present application has good prediction accuracy in face prediction, and the calculation complexity is low.

[0086] Therefore, the present application uses genomics features for model learning, and compared with the existing algorithm model technology mainly based on SNP site data associated with appearance features, the present application extracts a cleaner sample and improves the sample size, so that the model trained is more accurate in prediction.

[0087] Embodiment 3: As shown in the accompanying Figure 5 The present embodiment discloses a face prediction method, comprising:

[0088] Step S201, collecting genomics features of a user to be tested;

[0089] Step S202, inputting the genomics features of the user to be tested into a face prediction model to obtain predicted facial morphological features and obtain a face prediction result; wherein the face prediction model is a face prediction model constructed by a face prediction model construction method. Here, the face prediction model can be obtained by selecting from M pre-trained face prediction models, and the selection indicators include recognition accuracy, recognition speed, performance of the applied device, etc.

[0090] Embodiment 4: As shown in the accompanying Figure 6 The present embodiment discloses a face prediction model construction device, comprising:

[0091] A preprocessing unit acquires a training set, a test set and a validation set, wherein the training set, the test set and the validation set each include a plurality of genomics features and a plurality of real facial morphological features, and the plurality of genomics features and the plurality of real facial morphological features correspond one-to-one;

[0092] A modeling unit sets a training number, and trains a preset artificial neural network model multiple times according to the training number using the training set to obtain a plurality of face prediction standby models corresponding thereto;

[0093] A first evaluation unit uses the obtained plurality of face prediction standby models to predict the validation set to obtain corresponding loss function values, adjusts model parameters according to the loss function values in the validation set, selects optimal parameters, and outputs the optimal face prediction standby model as a face prediction model;

[0094] The loss function here is the sum of the weighted loss functions of three kinds of data, and the calculation formula is as follows: .

[0095] The three kinds of data are three kinds of representations of the predicted facial morphological features output by the model, i.e., the original high-dimensional vector, the principal component vector after principal component analysis, and the principal component vector after standardization.

[0096] a second evaluation unit, evaluating the face prediction model by using the test set and the validation set.

[0097] Herein the specific working process of the second evaluation unit includes:

[0098] 1. inputting the genomics features of each sample in the test set into the face prediction model, and outputting the predicted facial morphological features;

[0099] 2. determining the angle value between the predicted facial morphological features and the real facial morphological features;

[0100] Herein the angle value between the predicted facial morphological features and the real facial morphological features is determined, i.e. the SSA angle is calculated, and the calculation formula is as follows: .

[0101] 3. the validation set performs the same process as the test set;

[0102] 4. determining the mean value of all angle values corresponding to the test set and the validation set;

[0103] 5. evaluating the face prediction model by using the mean value. The evaluation can be setting a qualified condition, comparing the mean value with the qualified condition, judging whether the qualified condition is met, and in response to not meeting the qualified condition, adjusting the model parameters and training again.

[0104] Embodiment 5: as shown in the accompanying drawings, the present embodiment discloses a face prediction device, comprising: Figure 7

[0105] a collection processing unit, collecting the genomics features of a user to be tested;

[0106] a communication unit, inputting the target face data into the face prediction model;

[0107] a prediction processing module, obtaining a face prediction result by using the predicted facial morphological features; wherein the face prediction model is a face prediction model constructed by the face prediction model construction device.

[0108] It should be noted that: the face prediction model construction device and the face prediction device provided in the above embodiments are only exemplified by the division of the above functional modules, and in actual application, the above functions can be completed by different functional modules according to needs, i.e. the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0109] ​Embodiment 6: This embodiment discloses a face prediction device, comprising a processor, a memory, and one or more programs stored in the memory and configured to be executed by the processor, the program comprising instructions for performing steps in the face prediction method.

[0110] Here, the processor and the memory are connected by a bus, and the memory can include but is not limited to: a U disk, a read-only memory, a mobile hard disk, a magnetic or optical disk, and various media that can store computer programs.

[0111] Here, the face prediction device further comprises a transmission device and an input / output device, both of which are connected to the processor.

[0112] The above technical features constitute the best embodiment of the present application, which has strong adaptability and best implementation effect. Non-essential technical features can be added or reduced according to actual needs to meet the needs of different situations.

Claims

1. A method for constructing a facial prediction model, characterized in that, include: Obtain training, testing, and validation sets, where each sample in the training, testing, and validation sets includes genomic features and real facial morphological features; Set the number of training iterations, and use the training set to train the preset artificial neural network model multiple times according to the number of training iterations to obtain multiple corresponding backup facial prediction models. The obtained multiple alternative face prediction models are used to predict the validation set to obtain the corresponding loss function values. Based on the loss function values ​​in the validation set, the model parameters are adjusted, the optimal parameters are selected, and the face prediction model is output. The face prediction model was evaluated using a test set and a validation set; The pre-defined artificial neural network model includes an input layer, an output layer, and eight hidden layers between the input and output layers. The input layer obtains genomic features, denoted as X. ; The k-th hidden layer is , q represents the number of neurons in the hidden layer, and the q value is different for each hidden layer. The activation function in the network structure is the Tanh function as shown below; ; The value of the j-th neuron in the first layer of the hidden layer is: ; The value of the j-th neuron in the (k+1)-th hidden layer is: ; The output layer outputs the predicted facial morphological features, denoted as Y. The value of the j-th neuron in the output layer is: ; The facial prediction model was evaluated using a test set and a validation set, including: The genomic features of each sample in the test set are input into the facial prediction model, and the predicted facial morphological features are output. Determine the angle between the predicted facial morphological features and the actual facial morphological features; ； The validation set undergoes the same process as the test set. Determine the mean of all angle values ​​corresponding to the test set and validation set; The mean is compared with the qualification criteria to determine whether the qualification criteria are met. If they are not met, the model parameters are readjusted and the model is trained. Obtaining the training set, test set, and validation set includes: Multiple samples were identified and divided into training, testing and validation sets. Each sample included corresponding genetic SNP locus data and real facial morphological features. Principal component analysis was performed on all genetic SNPs in the training set to obtain a new feature space. 97% of the principal components in the new feature space of each sample were selected as the genomic features of that sample, and the true facial morphological features of each sample were obtained in the same way. All genetic SNP locus data in the test and validation sets are mapped to a new feature space in the training set. 97% of the principal components in each sample are selected as the genomic features of that sample, and the true facial morphological features of each sample are obtained in the same way. The steps for obtaining the true facial morphological features of each sample include: Step 1: Perform face scanning on the sample to obtain a 3D face image: Use a color handheld 3D scanner to obtain 3D face data. The scanning sequence starts from one ear of the subject, moves up and down in an S-shape through the nose and face to the other ear, then moves down to collect the image of the chin area that is covered, and finally returns to the starting position to complete the scan. The single frame curved surface range during the scanning process is 90mm×70mm to 180mm×140mm. Step 2: Obtain the original facial point set data corresponding to the 3D face image: Use the software built into the 3D scanner to process the 3D face image to obtain the original facial point set data; Step 3: Perform automatic non-rigid registration on the original facial point set data to obtain the corresponding real facial morphological features: The fully automatic non-rigid registration process is completed using FARS software and 3dMD patient software. If the quality of the 3D facial image of the sample is poor, the sample will be removed.

2. The facial prediction model construction method according to claim 1, characterized in that, The steps for obtaining SNP locus data for each sample include: Venous blood was extracted from the samples, and genetic SNP locus data were obtained through sequencing and typing. The genetic SNP locus data are preprocessed to output the genetic SNP locus data for each sample. The preprocessing includes cleaning and quality control.

3. A facial prediction method, characterized in that, include: Collect genomic characteristics of the users to be tested; The genomic features of the user to be tested are input into the facial prediction model to obtain the predicted facial morphological features and obtain the facial prediction result; wherein, the facial prediction model is a facial prediction model constructed using any one of the facial prediction model construction methods as described in claims 1 to 2.

4. A facial prediction model construction device, characterized in that, include: The preprocessing unit obtains a training set, a test set, and a validation set, each of which includes multiple genomic features and multiple real facial morphological features, and the multiple genomic features and multiple real facial morphological features correspond one-to-one. The modeling unit sets the number of training iterations and uses the training set to train the preset artificial neural network model multiple times based on the number of training iterations to obtain multiple corresponding backup facial prediction models. The pre-defined artificial neural network model includes an input layer, an output layer, and eight hidden layers between the input and output layers; The input layer obtains genomic features, denoted as X. ; The k-th hidden layer is , q represents the number of neurons in the hidden layer, and the q value is different for each hidden layer. The activation function in the network structure is the Tanh function as shown below: The value of the j-th neuron in the first layer of the hidden layer is: ; The value of the j-th neuron in the (k+1)-th hidden layer is: ; The output layer outputs the predicted facial morphological features, denoted as Y. The value of the j-th neuron in the output layer is: ; The first evaluation unit uses multiple prepared face prediction models to predict the validation set, obtains the corresponding loss function values, adjusts the model parameters based on the loss function values ​​in the validation set, selects the optimal parameters, and outputs the face prediction model. The second evaluation unit evaluates the facial prediction model using a test set and a validation set, including: The genomic features of each sample in the test set are input into the facial prediction model, and the predicted facial morphological features are output. Determine the angle between the predicted facial morphological features and the actual facial morphological features; ； The validation set undergoes the same process as the test set. Determine the mean of all angle values ​​corresponding to the test set and validation set; The mean is compared with the qualification criteria to determine whether the qualification criteria are met. If they are not met, the model parameters are readjusted and the model is trained. Obtaining the training set, test set, and validation set includes: Multiple samples were identified and divided into training, testing and validation sets. Each sample included corresponding genetic SNP locus data and real facial morphological features. Principal component analysis was performed on all genetic SNPs in the training set to obtain a new feature space. 97% of the principal components in the new feature space of each sample were selected as the genomic features of that sample, and the true facial morphological features of each sample were obtained in the same way. All genetic SNP locus data in the test and validation sets are mapped to a new feature space in the training set. 97% of the principal components in each sample are selected as the genomic features of that sample, and the true facial morphological features of each sample are obtained in the same way. The steps for obtaining the true facial morphological features of each sample include: Step 1: Perform face scanning on the sample to obtain a 3D face image: Use a color handheld 3D scanner to obtain 3D face data. The scanning sequence starts from one ear of the subject, moves up and down in an S-shape through the nose and face to the other ear, then moves down to collect the image of the chin area that is covered, and finally returns to the starting position to complete the scan. The single frame curved surface range during the scanning process is 90mm×70mm to 180mm×140mm. Step 2: Obtain the original facial point set data corresponding to the 3D face image: Use the software built into the 3D scanner to process the 3D face image to obtain the original facial point set data; Step 3: Perform automatic non-rigid registration on the original facial point set data to obtain the corresponding real facial morphological features: The fully automatic non-rigid registration process is completed using FARS software and 3dMD patient software. If the quality of the 3D facial image of the sample is poor, the sample will be removed.

5. A facial prediction device, characterized in that, include: The data acquisition and processing unit collects the genomic characteristics of the users to be tested. The communication unit inputs the target face data into the face prediction model; The prediction processing module uses predicted facial morphological features to obtain facial prediction results; wherein, the facial prediction model is the facial prediction model constructed by the facial prediction model construction device as described in claim 4.

6. A facial prediction device, characterized in that, It includes a processor, a memory, and one or more programs, said one or more programs being stored in the memory and configured to be executed by the processor, said programs including instructions for performing the steps of the method as described in claim 3.

Images